City, University of London Institutional Repository

Citation: Ortega, J. and Verdugo, G. (2015). The Impact of Immigration on the Local Labor Market Outcomes of Blue Collar Workers: Panel Data Evidence (15/07). London, UK: Department of Economics, City University London.

This is the published version of the paper.

This version of the publication may differ from the final published version.

Permanent repository link: http://openaccess.city.ac.uk/13838/

Link to published version: 15/07

Copyright and reuse: City Research Online aims to make research outputs of City, University of London available to a wider audience. Copyright and Moral Rights remain with the author(s) and/or copyright holders. URLs from City Research Online may be freely distributed and linked to.

City Research Online: http://openaccess.city.ac.uk/ publications@city.ac.uk

City Research Online

Department of Economics

The Impact of Immigration on the Local Labor Market Outcomes of Blue Collar Workers: Panel Data Evidence

Javier Ortega

City University London, CEP (LSE), CReAM and IZA

Gregory Verdugo

Banque de France

Department of Economics

Discussion Paper Series

No. 15/07

1

The Impact of Immigration on the Local Labor Market

Outcomes of Blue Collar Workers: Panel Data Evidence*

Javier Ortega†

City University London, CEP (LSE), CReAM and IZA

Gregory Verdugo‡

Banque de France and IZA

August 2015

Abstract

Using a large administrative French panel data set for 1976-2007, we examine how low-

educated immigration affects the wages, employment, occupations and locations of blue-collar

native workers. The natives in the sample are initially in occupations heterogeneous in the

presence of immigrants, which might reflect a different degree of competition with low-

educated immigrants. We first show that larger immigration inflows into locations are

accompanied by larger outflows of negatively selected natives from these locations. At the same

time, larger immigrant inflows into occupations come with larger outflows of positively

selected natives towards occupations with less routine tasks. While we find no negative impact

on employment, there is substantial evidence that immigration lowers the median annual wages

of natives. The estimated negative effects are also much larger in cross-section than in estimates

controlling for composition effect, which is consistent with the idea that endogenous changes

in occupation and location attenuate the impact of immigration on natives’ wages. We also find

much larger wage decreases for workers initially in non-tradable sectors and more particularly

in the construction sector, which are much less likely to upgrade their occupation or change

location in response to immigration inflows.

JEL : J15, J31.

Keyworkds : Immigration, Wages, Employment.

* We thank the INSEE for having made the data available. The Census data used in this paper are available upon request for researchers from

the CMH. The authors accessed the DADS data via the Centre d'Accès Sécurisé Distant (CASD), dedicated to the use of authorized researchers,

following the approval of the Comité français du secret statistique. The views expressed here do not necessarily reflect those of any of the

organizations with which the authors are affiliated. We thank Denis Fougère, Kyle Mangum, Manon Dos Santos, Ahmed Tritah and Muriel Roger and seminar participants at AMSE, Norface, SOLE, IZA-SOLE, CEP (LSE), and ESSLE-CEPR for very useful suggestions.

† Contact details: Department of Economics, School of Arts and Social Sciences, City University London, Northampton Square, UK-London EC1V 0HB, UK, email: javier.ortega.1@city.ac.uk

‡ Contact details: Banque de France, 31 rue Croix-des-petits-champs, 75049 Paris Cedex 01, France, email: gregory.verdugo@banque-france.fr

2

Introduction

A substantial part of the public concern about immigration in developed countries is a concern

about the impact that immigrants might have on the labor market outcomes of natives.

However, while much work has been done, credibly identifying the impact of immigration

still poses a significant empirical challenge.

A recent strand of the literature uses fixed and observable individual characteristics such

as education and experience to delineate the groups of natives and immigrants in competition

(see e.g. Borjas (2003)), or Aydemir and Borjas (2007)). However, recent research has

criticized the hypothesis of perfect substitution within education/experience groups. Indeed, if

immigrants and natives with similar education and experience levels are not directly in

competition, changes in immigrant supply may have little impact on native wages.4 As a

result, most of the impact of immigration could be concentrated on narrow groups of natives

who compete for jobs similar to the ones that immigrants tend to occupy.

To identify how much the impact of immigration varies across workers, a simple

empirical strategy would be to focus on natives who, before the immigration inflow, occupied

jobs where immigrants tend to be over-represented and are thus more likely to offer similar

skills in the labor market. However, such an approach faces important empirical challenges.

Indeed, natives initially occupying those jobs might endogenously respond to immigrant

inflows by changing their occupation or location. As a result, cross-sectional changes within

groups will be affected by compositional changes in the characteristics of workers.5 Because

4See Ottaviano and Peri (2012) and Manacorda et al. (2012) for evidence that immigrants and natives might be imperfect substitutes within education/experience cells in respectively the US and the UK. Peri and Sparber (2009) show that low-skill natives in local labor markets

receiving more immigrant inflows tend to specialize in occupations requiring more abstract tasks in response to immigration. Dustmann et al.

(2013) show that recent immigrants start working in occupations offering a much lower wage than natives with similar observable characteristics.

5Using US decennial data, Card (2001) and Cortes (2008) have found no evidence of native outflows in response to immigrant inflows while Borjas (2006), on the other hand, reports strong displacement effect. More recently, using US annual aggregate data, Wozniak and Murray (2012) find that immigrant’ inflows are correlated with declines in outflows of low skill natives in the shorter run of one year. Recent

European studies found stronger evidence of displacement: using Italian data, Mocetti and Porello (2010) find evidence of displacement of

low skilled natives following immigrant inflows. For the UK, Hatton and Tani (2005) find consistently negative displacement effects.

3

job changers might not be selected randomly from the sending population, it is challenging to

identify how immigration affects the outcome of these workers if the composition of workers

across occupations changes significantly.

In this paper, we revisit the effect of low-educated immigration on the local labor

market opportunities of natives by exploiting detailed information on individual labor market

trajectories from a very large administrative panel data of the French labor force over a period

of 30 years. This large panel data provides for about 4% of French private sector employees

exhaustive information on the wages, occupations, the number of days of work, and the

geographical location at the municipality level of each job held during the period 1976-2007.

We also use very large (25%) sample extracts from the Census to estimate changes in low-

educated immigrant inflows across locations and to construct our instrument for immigrant

inflows. We combine these two datasets to investigate how the labor market outcomes of

blue-collar natives respond to inflows of low-educated immigrants.

Using panel data in this setting is attractive for several reasons. A first key advantage is

that detailed information on individual labor market trajectories is available. This implies we

can follow groups of workers narrowly defined by their initial occupation and test for

potential heterogeneity. In addition, given we would expect natives initially in jobs with many

low-educated immigrants to be more affected by immigration, the econometric model is

estimated separately for natives initially occupying blue-collar jobs across industries

heterogeneous in their initial share of immigrants.

A second advantage of using panel data is that we are able to control for unobserved

heterogeneity of workers. With longitudinal data, we can isolate the causal effect of

immigration on wages from any compositional change. In addition, we can directly

4

investigate how immigration affects the selection patterns of natives across occupations or

locations.

In the first part of the paper, we present some new facts on the effect of immigrant

inflows on the selection of natives across locations and occupations. We find compelling

evidence in alternative datasets of a moderate positive correlation between low-educated

immigrant inflows and outflows of blue collar natives from the location. Quantitatively,

baseline 2SLS estimates suggest that a 10 p.p. increase in the immigration rate (defined as the

ratio between the number of low-educated immigrants and blue collar natives in the

commuting zone) increases the outflow rate of blue collar natives by 3.2 p.p. An important

result is that outflow rates vary dramatically across occupations. In particular, our IV

estimates indicate a much larger displacement effect for workers initially in the most

immigrant-intensive industries such as non-tradable industries and particularly the

construction sector.

Second, consistent with evidence from Peri and Sparber (2009) for the U.S. or Ortega

and Verdugo (2014) for France, we find that natives are more likely to change occupation

following immigrant inflows. In particular, our results suggest they tend to move to

occupations of better quality which require less routine tasks. However, once again, the

estimated effect varies importantly across occupational groups. In particular, we do not find

any evidence of occupational upgrading for low-skill workers in the construction sector or in

the non-tradable sector. This last result suggests that a substantial share of workers,

particularly those with the lowest skill levels are not able to protect themselves from

competition with immigrants by moving to other occupations.

Third, there is also strong evidence that, within groups, workers changing location or

occupation are not a random sample of the sending population. Specifically, workers moving

to occupations with less routine tasks tend to be positively selected, in the sense that they

5

initially have higher wages conditional on their location and initial occupation. In contrast,

workers changing location tend to be negatively selected.

In the second part of the paper, we examine the impact of immigration on the wages and

employment of natives. We use variations from a balanced sample to isolate the impact of

changes in composition from the impact of immigration on wages. We find no evidence of a

negative impact of immigration on the average number of days worked or the employment

rate. In contrast, we find immigration to be negatively correlated with median annual wages,

the effect being the largest for workers in the non-tradable sector, particularly those in the

construction sector. For this group, our estimates suggest that an increase in 10 p.p. in the

immigration ratio at the local level generates a decrease of 3.6 log points in the median annual

wage. We also show that the equivalent cross-sectional estimates –i.e. the estimates with the

same data when their longitudinal dimension is not exploited- systematically overestimate the

impact of immigration on wages.

Overall, these results suggest that the impact of immigration is heterogeneous both

across and within occupational groups. The selective reallocation of a substantial share of

natives to different occupations and locations attenuates the final effects on wages, as argued

by Peri and Sparber (2009), but the extent of this reallocation varies widely across groups of

workers. In particular, there is no evidence of occupational upgrading for low skilled workers

in the construction sector. Finally, there is also strong evidence that within groups workers

moving to better occupations tend to be initially positively selected which implies that the

negative effect on wages is concentrated on workers with the lowest wages within groups.

There is a growing but still relatively small literature on the impact of immigration on

natives’ labor market outcomes using panel data. Most of the existing papers (see in particular

De New and Zimmermann, 1994, Bratsberg and Raaum, 2012, and Bratsberg, Raaum, Røed,

and Schøne, 2014) do not exploit the geographical variation in the number of immigrants and

6

find the effect of immigration on wages to be heterogeneous across different groups of natives

depending on their degree of complementarity/substitutability with respect to immigrants.

Foged and Peri (2014) considers instead the geographical and cross industry variation in the

proportion of immigrants in Denmark, and shows that immigration has a positive wage impact

on the less skilled natives and encourages them to work in more complex occupations. With

respect to this literature, the main contribution of our paper is to identify the nature of the self-

selection of natives across cities and occupations following immigration and also to

understand the extent of the bias incurred when using cross sectional data instead of panel

data to assess the impact of immigration.6

The remainder of the paper is organized as follows. The first section presents the data

and provides some descriptive evidence on immigration into France. The second section

discusses the empirical framework. The third section investigates the relationship between

native locations and occupations and immigrant inflows. The fourth section examines the

impact of immigration on employment and wages. The last section concludes.

I) Data and descriptive evidence

Data Sources

Our primary source of data for the analysis comes from the matched employer-employee

panel DADS (in French Déclaration Annuelle de Données Sociales) collected by the French

National Institute for Statistics (INSEE).7 The sample contains earning histories for all

individuals born in even-numbered years in October. Annual DADS data are available from

1976 to 2007 except for 1981, 1983 and 1990 were the data were not collected.

6 See also Lull (2014) for an interesting evaluation of compositional changes in the native population following immigration using a structural econometric approach.

7 See e.g. Abowd et al. (1999) or Combes et al. (2008) for recent examples of papers using this dataset.

7

Three features of this dataset make it well-suited for studying the impact of

immigration: first, DADS data are collected from compulsory fiscal declarations made

annually by all employers for each worker and are thus considered very reliable.8 The annual

wage data is considered of very good quality: the reporting, made by the employer, is used to

compute the income tax of the worker. Employers have no incentives to misreport wages as

this is severely punished with fines. Second, DADS data being an administrative panel data

collected for fiscal purposes, involuntary attrition has been evaluated to be modest.9 Most of

the attrition comes either from an exit from a sector covered by the DADS or a supply of zero

days of work in a given year. Third, the sampling size is very large: we have information on

wages for 350,000 individuals per year over the period, representing about 4% of the

population working in the private sector.10

The data contains a unique record for each employee-establishment-year combination.

For each individual job spell of any length in a given firm, the DADS collects information on

earnings, whether the job was part or full-time, the number of days of work and the location at

the municipality level. One drawback is that information on the number of days of work and

the precise number of hours worked appears to be rather noisy.11 A relatively large share of

workers is reported to have worked full-time full year but have wages well below the

minimum wage. This creates a limitation to evaluate daily wages or changes in number of

days worked.

8 Not all the sectors of the economy are covered each year and the degree of coverage increases over time. In particular, civil servants and most large public sector firms are excluded until the 1990s. Using LFS data, we estimate that they represented approximately 8% of the labor

force during the 1980s. 9 Koubi and Roux (2004) document that most of the temporary attrition from the DADS panel corresponds in practice to inactivity or a work

outside of the DADS covered sector (such as self-employment, or work in the public sector until the 1990s). Attrition in the DADS panel has also been shown to be much lower than in typical survey-based panels such as the European Community Household Panel (ECHP)( Royer

(2007) ).

10 The sampling size doubles in 2002 when individuals born in odd-numbered years in October are added to the sample.

11 Information on whether an employment spell was full or part time is available over the entire period but the number of hours worked is only available after 1993 (see Aeberhardt et al. (2011) for a discussion). Following the current practice, we have chosen not to use it.

8

We aggregate each job spell to obtain the total annual income and number of days of

work within a year. We retain information on occupation and industry of the job held during

the largest number of days. Note that education is missing from the data. Another important

point is that there is no information on nationality in the DADS but the data indicate whether

an individual is born abroad. We define as natives, in this dataset only, individuals who are

born in France and exclude individuals who are born abroad from the native sample.12

Because of this last limitation, we do not rely on DADS data to estimate changes in the

number of immigrants across local labor markets over time. The lack of information on the

country of origin makes it impossible to construct an instrumental variable for changes in the

immigration ratio using differences in settlement patterns across immigrant groups. Instead,

we rely on Census data to estimate the changes in the number of low-educated immigrants

across commuting zones. Censuses of the population took place in 1975, 1982, 1990, 1999

and 2007. An important advantage of this dataset is that we use 25% extracts (20% in 1975)

of the Census population to compute changes in the immigrant ratio across locations over

time. Such large sample size are essential for an analysis of the impact of immigration since it

renders the results immune from attenuation biases as identified in Aydemir and Borjas

(2011). As is conventional, an immigrant is defined as a foreign-born individual who is a non-

citizen or naturalized French citizen.

Local labor markets are defined using the 2010 definition of commuting zones (zones

d’emploi). Commuting zones are designed by the INSEE to approximate local labor markets

12 Many French-born citizens who should not be counted as immigrants were born in Algeria before independence in 1962: using the census, their share among 18-65 years old natives is 2.2% in 1982 and 1% of in 2007. More generally, the share among natives of French-born citizen who are born abroad is rather small and declining over time: 4.4% and 3.2% in respectively 1982 and 2007. Since we are not able to

distinguish them from immigrants, they are excluded from the DADS sample of natives.

9

using information on daily commuting patterns. They aggregate the 36 699 existing French

municipalities into 297 labor market regions.13

Our empirical implementation uses variations in low-educated immigrant inflows across

commuting zones in France from 1975 to 2007. We estimate first-differenced models in

which we relate changes in native labor market outcomes obtained from the DADS data with

changes in the share of low-educated migrants obtained from the Census data using years in

which both census and DADS data are available.14 Low-educated immigrants are defined as

immigrants with a level of education below high-school graduation. Since DADS data does

not contain information on individuals out of the labor force, we focus on prime-aged male

workers aged more than 25 and less than 54 who have relatively strong labor market

attachment and for whom non participation during a full year is less likely to be a major

issue.15 This implies we concentrate on individuals aged 25 to 45 in census year t and 32 to 52

or 34 to 54 in census year t+1, where t is a census year and t+1 the year of the next census,

and that we consider changes in the number of immigrants within commuting zones over

periods of 7 to 9 years.

Immigration in France: Descriptive Evidence

According to the last census, in 2007, 5.2 million immigrants lived in France, which amounts

to 8.3% of the population. The share of immigrants in the population is thus lower than in the

U.S. and the U.K. (respectively 11.5% and 11.9%, see Dustmann et al., 2013, p. 11). However,

13 Commuting zones are also used with the DADS data by Combes et al. (2008) and Combes et al. (2012). They are defined in a consistent way over time. We drop commuting zones from Corsica (less than 0.3% of the population), as a change in the département code in 1976

complicates their matching across datasets over time.

14 Given DADS data were not collected in 1975 and 1990, we match census data from the 1975 and 1990 census with respectively the DADS

data from 1976 and 1991.

15 We also apply these restrictions to avoid issues with changes in retirement age over time. Young workers are also eliminated to avoid problems with potentially endogenous labor market participation in case immigration influences education decisions (see Hunt (2012) or

their employment probability (see Smith (2012).

10

from 1975 to 2007, France experienced an increase of 5 p.p. from 13% in 1975 to 18% in 2007

in the share of immigrants among the group of male workers16 with a level of education below

high-school. The geographical origin of immigrants also changed during the period: the share

of European immigrants decreased from about 60% in 1975 to only 32% in 2007.

Table 1 reports the share of foreign born workers among blue collar workers in 1999 in

the tradable and non-tradable sectors for the country as a whole and for some large cities (see

Appendix for details on industries and occupation classifications used in the paper).17 As shown

in the table, low-educated immigrants tend to be overrepresented in some sectors and regions,

particularly in the non-tradable sector. The share of foreign born workers is 4 p.p. and 10 p.p.

higher in respectively the non-tradable sector and the Construction sector relative to the tradable

sector. These figures suggest that competition for jobs with low-educated immigrants may be

strongest in the non-tradable and construction sector.

Immigrants are also unevenly distributed across regions: while only 3% of blue collar

workers are foreign born in Brittany in the non-tradable sector, the share of foreign born is 33%

in Paris. Similarly, the share of foreign born blue collar construction workers is 45% in Paris

compared to 5% in Brittany. However, Table 2 indicates that in both regions, the share of

foreign born workers expanded in the Construction sector in the last 30 years.

II) Theoretical Framework

Immigration may impact the labor market outcomes of natives through several channels which

are interrelated, including wages, location or occupations. Here we discuss these channels.

Assuming a CES production function with different occupation groups each of them

16 Unless otherwise indicated, figures in this section male workers aged 18-64 which are not students or in the military.

17 We rely on standard classification systems of industries. See appendix for details. Following Hanson and Slaughter (2002) and

, the group of tradable industries includes manufacturing, agriculture, mining, finance and real estate.

11

aggregating individuals heterogeneous in their labor productivity, the period t log wage for

individual i in local labor market l and in occupation group k can be written as (see Appendix):

log ( , ) log k ltit klt it i

klt

Iw k l B X

N

, (1)

where is the elasticity of substitution across occupation groups, i is individual i’s

unobserved (and constant) productivity, itX is a set of individual observable characteristics,

ltN is the number of natives in occupation group k and ltI the number of low-educated

immigrants in location l . We discuss below how we empirically define these occupation

groups. A key point is that, as in Smith (2012) or Dustmann et al. (2013), we do not pre-

allocate low-educated immigrants to a particular group but estimate the response of various

native groups to a change in the share of low-educated migrants in the location. To introduce

heterogeneity in the impact of immigration across groups in the simplest way, we follow

Dustmann et al. (2013) and assume that a share k of low-educated immigrants have a skill

level corresponding to the occupation group k , and that they are perfect substitutes within

occupation groups.

Workers might endogenously adjust to immigration by changing location as argued by

Borjas et al. (1997) or occupation as argued by Peri and Sparber (2009) and Amuedo-

Dorantes and de la Rica (2011). If immigration changes the relative price of skills in the labor

market, some workers might move to another location or to another occupation in response to

immigrant inflows. To illustrate in a simple way how endogenous self-selection may

confound the impact of immigration in cross-section regression in this framework, assume

workers differ in their ability to move across locations as in Moretti (2011) or Beaudry et al.

(2010) or in their ability to move across occupations. As a result, individuals changing

location or occupation are not a random sample of the initial population of workers. Assume

next that native outflows are such that the efficiency units of labor supplied by natives change

12

by klt k k ltN I when the share of immigrant increases by ltI in the location.18 The

parameter k is the share of native net outflows in group k with respect to a change in

immigrant labor supply. When 1k , there is perfect displacement, while when 0k there is

no response. This implies that changes in the average log wage log kltw of natives in

occupation group k , location l and period t can be expressed as:

1log logklt klt k lt klt klt kltw B p X (2)

where ,

1klt i

i k lkltN

is the average productivity of workers and ,

1klt it

i k lklt

X XN

is the

average individual observable characteristic. The term ltp captures the change in the share

of low-educated migrants in the location while the term 1klt klt captures changes in the

unobserved productivity of workers in occupation group k . There is positive selection

correlated with immigrant inflows if 1 , 0klt klt itcov p and negative selection

otherwise.

In this simple framework, the parameter (1 )kk k

is a function of the elasticity of

wages to the labor supply of immigrants, and of the elasticity of mobility of natives. If

mobility costs are sufficiently low for a large number of individuals, that is 1k , native

outflows will equalize wages across locations and thus immigration does not have an impact

on the local wages of natives, but only at the national level, as in Borjas (2006). Instead, if

mobility costs are substantial, native internal mobility might not be sufficient to offset the

local effect of immigrant inflows on wages. To evaluate the importance of such channels, we

18 The previous equation is obviously a reduced form. Modeling the sorting of workers across locations is beyond the scope of this paper.

13

test for the impact of immigration on location and occupation and investigate the patterns of

selection of location and occupation-movers.

Econometric Model

To take the previous equations to the data, we need to make some additional assumptions. We

assume, as common in the literature, that changes in log kltB over time in a given location and

occupation can be decomposed by a full set of fixed effects. Then, equation (A) and (B) lead

to simple regression models of the form:

klt k lt k lt k klt kt kr klty p Z X (3)

where klty is the change in a given outcome between two periods for occupation group k in

location l , kt are time fixed effects and kr are region fixed effects. The vector ltZ contains

several location and industry specific factors varying over time. Following the literature,19 the

term ltp in the empirical model is defined empirically as the change in the low-educated

immigrant ratio with respect to the initial number of blue collar workers in the location:

1

1

,lt lt

lt ltlt I I

l tN

I Ip e

. The use of a similar numerator across occupation groups facilitates the

interpretation of the results given the size of groups might vary widely. The term 1lt ltI Ie is an

indicator function which is equal to one if the number of low-educated migrants is strictly

increasing in the location and is zero otherwise. We condition the immigration rate to be

positive to avoid our results to be affected by the rare locations and periods in which the

numbers of low-educated migrants decrease in the population.

The model is estimated by pooling multiple decades as stacked first differences.

Because the model is estimated in differences, it eliminates time-invariant wage differences

19 See e.g. Card (2001), Card and DiNardo (2000), or Mazzolari and Neumark (2012).

14

across occupation groups and locations that may be correlated with the share of immigrants.

The specification also includes changes in average individual-level demographic controls (

kltX ) and changes in area level controls as well ( kltZ ). The additional controls included in the

regressions are the changes in the share of white collar and blue collar workers, the share of

workers in construction, the overall share of workers in manufacturing industries and the

average age of workers. The model also includes regional fixed effects. As in

or Smith (2012), regressions are weighted by the number of observations used to

compute the dependent variable: this implies that we weight first-differenced equations by

1/2

1(1/ 1/ )klt kltN N

where lktN is the number of observations used to compute the outcome

variable.20

Native Groups’ Definition

Equation (2) makes clear that, absent a strategy for isolating variations in wages that are

independent of changes in the average unobserved i in the occupation, changes in wages

reflect both the impact of immigration on the supply of labor and on the unobserved average

productivity of workers. To get rid of the change in unobserved characteristics of workers, we

adopt a simple empirical strategy. The panel aspect of our dataset allows us to define the

‘treated’ occupation group of natives by their initial occupation and location.

Using the initial occupation to define occupation groups has several crucial

advantages. Natives initially sharing the same occupation are more likely to offer in the labor

market a more similar set of skills. Our rich dataset allows us to focus on narrow groups of

workers who are more likely to compete with low-educated migrants, namely those in

20 This formula comes from straightforward calculations of the variance of a first-difference variable measured with errors under the assumption that the measurement error is proportional to the number of observations and is independent across years.

15

occupations with a larger share of immigrants. If the effect of immigration on wages differs

importantly across groups of natives, it might be important to allow for a different effect.

We also control directly for changes in unobserved heterogeneity through the use of a

balanced sample in which the composition of natives included in the sample is maintained

constant over time. A second interest of our approach is that our strategy controls directly for

changes in unobserved heterogeneity through the use of a balanced sample. Estimates of the

impact of immigration obtained from a balanced sample are by definition not driven by an

endogenous change in the composition of natives in the occupation group.

Occupation groups are defined here by using the interaction between being a blue

collar worker and working in a given industry in the initial period. We use four different

groups. We define a first group pooling all blue collar workers to estimate the average effect

of immigration on these workers. Two other groups are defined by distinguishing between

workers in the tradable and non-tradable industries. Finally, we define a fourth group isolating

workers from the construction sector from the non-tradable industries group. If the supply

effect of immigration differs across groups of natives, we should expect a larger effect on

workers initially in non-tradable industries, particularly in the construction sector. On the

other hand, if blue collar workers are all perfect substitutes in production, then we expect the

impact of immigration to be similar across groups.

Identification

As discussed previously, it is very unlikely that immigrants’ geographic settlement decisions

are exogenous to local labor market conditions. If immigrants settle disproportionately in

areas with better local labor market conditions, then ordinary least squares (OLS) estimates of

k will be biased. One important concern is the possibility that pre-existing trends are

correlated with both changes in the immigrant ratio and changes in the variables of interest. If

this is the case, the estimates presented here may simply be a spurious correlation. To deal

16

with this issue, the model includes a control for regional specific trends r for 22 French

regions. Due to the inclusion of a vector of regional dummies kr , the coefficient of interest

k is identified by within regional variation. The inclusion of regional dummies means that

any confounding factor would have to vary within region over time. Including these fixed

effects thus addresses some of the concerns raised by Borjas et al. (1997) when one does not

control for the various confounding factors affecting outcomes across locations.

As in Card (2001) and Cortes (2008), our identification strategy uses the initial

proportion of co-nationals in the commuting zone as an instrument for future immigrant

inflows. Specifically, the predicted number of low-skill immigrants in region r is given for

each census year t by

, 1

1

, 1

ˆ *cl t

lt ct ct ct

c cc t

II I I

I

where , 1

, 1

cl t

c

cl

t

I

I

is the proportion in the previous census date 1t of country c immigrants,

including both low and high skill immigrants, living in region l , while ctI is the total number

of immigrants from country c in France in year t . Given the large sample size of the census,

we distinguish groups of immigrants by using the maximum number of nationalities available,

namely the 54 different countries of birth which are always reported separately across

censuses. Following Hunt and Gauthier-Loiselle (2010), we explicitly determine 1ct using

immigrants from all education and experience levels to have a greater role of geography and

ethnic networks. By doing so, our aim is to give less importance to economic factors that

might attract workers with low levels of education and experience specifically in a given

region. Because the endogenous variable is a percentage, we define our final instrument by

17

using the change in the number of predicted immigrants in the location divided by the initial

number of natives, to define our final instrument as: 1

1ˆ ˆ

lt lt

lt ltI I

lt

I I

Le

The validity of the instruments used to predict changes in the immigrant ratio over

time is examined in Table 3. Observations correspond to changes between census years, i.e.,

1975-1981, 1981-1990, 1990-1999, and 1999-2007.

Column (1) reports estimates from a simple bivariate model while column (2) includes

a full set a control variables. In both specifications, the coefficient is positive and strongly

significant. A comparison between columns (1) and (2) indicates that adding the control

variables lowers by a third the estimated parameter but also raises the precision of the

estimate. In column (3), we examine results from unweighted estimates: the coefficient

declines by a fourth but still remains statistically significant. Overall, the Fisher statistics of

the instrument indicate it is reasonably strong across the various specifications. With F-

statistics greater than 10 in most specifications, they easily pass the weak instrument test.

III) Immigrant Inflows and Natives’ Mobility Patterns

Before investigating the impact of immigration on wages and employment, we first provide

evidence on the relationship between immigrant inflows and natives’ location and

occupations. In contrast with the existing literature, the panel dimension of the data allows us

to focus on natives defined by their initial occupation and to investigate selection patterns. In

a first and a second subsection, we investigate the correlation between immigrant inflows and

changes in locations and occupation of natives. In a third subsection, we look at the selection

patterns of movers in an attempt to understand how selective change in location and

occupation affects the composition of natives labor force within locations and occupations.

Local labor market mobility

18

We begin by assessing the relationship between local immigrant inflows and native inflows

and outflows. A simple accounting identity relates the net annual change in native total

population kltN of occupation group k in location l with the number of individuals who

moved into the location ( kltI ) and the number of individuals who left the location ( kltO ):

1 1 1klt klt kltN I O .21 Following Card (2001), Card (2009) or Cortes (2008), we estimate

separately for each occupation group k the model of Eq. (3) in which the dependent variable is

either the inflow rate 1 /k

lt kltI N or the outflow rate 1 /k

lt kltO N .

Panel 1 in Table 4 shows the results for different groups of industries. Within each

panel, the first line provides OLS results while the second line reports 2SLS results. For all

groups of workers, with the exception of blue collar workers in the tradable sector, both IV

and OLS results indicate there is a positive correlation between immigrant inflows and native

outflows in the initial location. OLS results indicate that an increase of 10 p.p. of the

immigration rate into the location is associated with an increase in outflow rates of 0.7 to

0.9 p.p. depending on the group. On the other hand, IV estimates are up to four times larger

than OLS estimates. Interestingly, the estimated effects are much larger for blue collar

workers initially in immigrant intensive sectors such as those in the non-tradable sectors and

in the construction sector: we find that an increase of 10 p.p. in the immigration rate raises the

share of movers by 1.6 p.p. for blue collar workers and by 3.6 p.p. for workers in the

construction sector.

Turning now to the relationship between variations in the immigrant ratio and native

inflows, OLS results indicate for all occupation groups a strong positive correlation.

Parameter estimates are remarkably similar across groups of industries. The OLS estimates

21 Outflows are computed by using information on the occupation of the individual in the period t+1, and whether this individual has changed location in t, independently of her occupation in period t. Inflows are computed by using the number of individuals in the

occupation in period t+1 who worked in a different location in period t independently of their initial occupation.

19

suggest that an increase in 10 p.p. of the immigration rate is correlated with an increase of

1.7 p.p. in the inflow rate of natives into these occupations. These parameter estimates imply

that the arrival of 100 immigrants into the location is correlated with the exit of 16 native blue

collar workers and the entry of about 15 native blue collar workers.22 However, there is no

strong evidence for a causal effect on native inflows. While 2SLS estimates are positive and

not very different from OLS estimates, they are measured very imprecisely and are not

significantly different from zero.

The previous results have several limitations related to the characteristics of the DADS

data. The sample only includes individuals with a positive number of hours worked in the

private sector in both periods to compute inflows and outflows. Selective attrition to

nonparticipation or to a sector not covered by the DADS data could bias our results if

immigration is correlated with a large share of native workers dropping out of the labor

market or moving to the public sector. To address this concern, we assess the robustness of

the previous results by using alternative inflows and outflows rates computed with the French

census. An important advantage of the Census data is that it contains the entire population and

also includes retrospective information of the location at the municipality level at the time of

the previous census. Unlike in the DADS data, information on the initial occupation of native

workers is not available in the Census. Instead, we define groups by using information on

education and use information on the previous location to define inflows and outflows rates

across commuting zones for different education groups. We use four education groups: two

low skilled groups, primary or secondary education, and high-school and university

graduates.23 To be able to make a comparison with the previous estimates, our dependent

22 When the model is estimated using blue collar workers, the native outflow rate and the immigrant inflow rate are both divided by the initial number of blue collar workers in the location.

23 See the Appendix for details on the construction of these education groups.

20

variable is, as previously, the change in the share of low-educated migrants over the initial

number of native blue collar workers.

Consistent with the previous evidence, Census data estimates in panel 2 of Table 4

strongly indicate a positive correlation between both outflows and inflows for low skilled

workers. OLS estimates are slightly lower than those obtained with DADS data, indicating an

increase in 0.4 p.p and 1.1 p.p. for respectively inflows and outflows for an increase in the

immigrant ratio of 10 p.p. As previously, 2SLS estimates are only statistically significant for

outflows. The estimated IV coefficients are also much larger than the corresponding OLS

estimates, indicating an increase of about 2.4 p.p. and 2.7 p.p. of the share of movers for

respectively primary and secondary education workers for an increase of 10 p.p. in

immigration rate.

Taking the previous estimates together, three things are clear. First, immigrant inflows

appear to be positively correlated with both larger inflows and outflows of native blue collar

workers and of low skilled natives across employment areas. The evidence also suggests that

only native outflows seem to be causally related to immigrant inflows. Second, the fact that

there are two opposite inflows and outflows indicates that immigration is correlated with a

change in the composition of native blue collar workers in the location. Our results point to

the evidence of a much stronger displacement effect on native workers initially in jobs more

likely to be taken by low-educated immigrants. Third, the fact that immigrant and native

inflows are also positively correlated indicates that common positive economic shocks might

drive both native and immigrant location choice. This correlation between immigrant location

choice and local economic conditions should bias estimates of the impact of immigration on

labor market outcomes of natives.

21

Occupational Mobility

Next, we examine the impact of immigration on the occupations of natives. Following the

literature, we do not examine whether immigration impacts the probability to change

occupation but whether natives are more likely to move to better quality occupations in

locations with larger immigrant inflows. To capture changes in the skills supplied across

occupations over time, we focus on changes in the average routine to abstract intensity of

tasks performed in the occupations.24 The task contents of an occupation provides an

approximation of the basic skills required to perform it ( Autor et al. (2003),Acemoglu and

Autor (2011),Goos and Manning (2007) ). To interpret the parameter estimates, our routine to

abstract intensity index variable is normalized to have an average of zero and a standard

deviation of one across the distribution of occupations. Note that, in the initial period, blue

collar workers can be in one of the 6 distinct occupations that we have in our classification.25

Within the blue collar workers group, the occupation with the lowest routine to abstract skill

intensity is “laborer” with an index of 0.51 while “machine operators” have an index of 1.30.

In the final period, there is no restriction and workers initially in the blue collar worker group

may be in any kind of occupation.

Table 5 shows the results. Within each panel, the first column provides intent to treat

estimates using all workers initially in the occupation group, including those who have moved

to another location or occupation. In the second column, those moving from the location have

been excluded while the third column also excludes those who are not in the same occupation

group. Finally, the last column uses the variations from repeated cross-section in the occupation

group and location. Each subpanel refers to a different occupation group.

24 Abstract tasks are "complex problem solving" while routine tasks require repetitive strength and motion and non-complex cognitive skills and thus do not require good language skills. Data on task intensity come from the abstract and routine task intensity indexes calculated by

Goos et al. (2010, Table 4 p.49) from the Occupational Information Network (ONET) database that we have matched manually with French occupations classifications. See Appendix for details.

25 See Appendix for details.

22

In both samples, OLS results indicate very small, slightly positive coefficients, which

are most of the time statistically insignificant. On the other hand, 2SLS models indicate that

there is clear evidence that immigration is correlated with a decrease in average routine

intensity for native workers. Quantitatively, we find that an increase of 10 p.p. of the

immigrant ratio lowers the average routine to abstract intensity by 7.9 p.p. and 3.9 p.p. for

respectively workers initially in the tradable and non-tradable sector. In contrast, for

construction workers, there is no evidence of a correlation between immigration and the

change in routine intensity levels: the coefficient is much lower than for other groups and is

not statistically significant.

We also find very little difference between estimates including and excluding location-

movers in column 1 and 2. We also observe no effect of a much smaller effect when we

concentrate on those who stay in the same occupation group during both periods in Column 3.

Note that by definition these workers can only have moved to an occupation within the blue

collar worker group. The coefficient is also small and not statistically significant for most

occupation group with the cross-section sample. These results imply that most of the effect is

driven by workers who have moved to an occupation outside of their initial occupation group.

Is there Positive or Negative selection in Change in Location or Occupation?

We now investigate how individuals changing occupation or location are selected with respect

to the sending population. To do so, we first estimate the residual wage dispersion within

occupation group and locations by regressing the individual log daily wages for each occupation

group in each year against commuting zone fixed effects and on a full set of age fixed effects.

By using a residual wage dispersion with respect to the sending population, we investigate

whether those who have moved or those who have left had lower or higher wages than those

who had stayed with respect to the wage distribution in the initial period. Then, following

23

Borjas (1999), we define positive selection for occupation group k initially in location l in

period 1t as a situation in which:

(log | movers in t+1) (log | stayers in t+1)iklt ikltE rw E rw

where ikltrw is the residual wage level in the original location in the initial period. If there is

positive (resp. negative) selection, emigrants are on average more (resp. less) productive than

non-migrants with respect to the distribution of wages in the initial location. To investigate

these selection patterns, we first run the following regressions at the individual level:

1 2

1 ( )iklt k iklt k iklt lt k lt k lt k klt kt kr kltMove rw rw p p Z X

where the dependent variable 1ikltMove takes the value 1 if the individual has left the location

in period t+1. The coefficients of interest are 1

k and 2

k , respectively the coefficients of the

residual wage in the initial period and of an interaction term between the change in immigrant

ratio and the initial residual wage. The first coefficient indicates whether those who had moved

to another commuting zone had lower or larger relative wage with respect to the initial wage

distribution in the group in their initial location. The second coefficient tests for a potential

interaction between the selection term and inflows of immigrants in the location. Estimation is

based on 2SLS using the previously described instrument for ltp and the interaction of this

instrument with the residual wage for the interaction term iklt ltrw p .

Results are presented in Panel A of Table 6. Column 1 shows that movers in the

occupation group of blue collar workers are negatively selected. Parameter estimates indicate

that an increase of one standard deviation of the residual wage (about 0.32), decreases the

probability to change location by 3.7 p.p. (0.32 x 0.117). In Column 2, we introduce an

interaction term between the residual wage and the share of immigrant inflows. The term is

small and statistically insignificant. This suggests that there is no evidence that negative

24

selection varies in places receiving more or less immigrants. Other columns show similar

estimates for other occupation groups: for all groups, we also find a negative selection with

respect to the sending population and no effect of immigrant inflows on the selection patterns.

In Panel B, we estimate a similar model but in which the dependent variable is the change

in routine task intensity in the occupation between period t+1 and t for a given worker. In order

to guarantee that the variations in occupations come from individuals experiencing the

immigrant inflows and thus who have stayed in the location, we focus on location stayers in

this sample. Across occupation groups, we obtain a negative coefficient of the residual wage:

this implies that individuals with higher initial wages are more likely to experience a decrease

in their routine intensity during the period. Thus, individuals moving to occupations with more

abstract tasks and less routine tasks are positively selected. As previously, the interaction term

is not statistically significant which implies there is no strong evidence that the selection pattern

varies widely depending on immigrant inflows received by the location.26

Overall, the results presented in this section confirm the intuition from previous work

that changes in locations and occupations are endogenously related to immigrant inflows.

Especially important is the fact that the impact differs across occupation groups, with no

evidence of an impact of immigration on occupations and larger displacement effects for

construction workers.

By studying the selection patterns of workers, we were also able to highlight that those

who changed occupation and location are not a random sample of the initial population: workers

who change location tend to be negatively selected and have lower initial wages with respect

to the sending population. This implies that the selective exit of workers with lowest initial

wage from the location will tend to increase wages in the location in the second period. In

26 We explored more flexible specifications to estimate the interaction term and also estimated the previous models using OLS. The results were broadly similar and are available upon request.

25

contrast, workers who move to occupations with less routine tasks tend to have higher initial

wages and thus tend to be positively selected. This outflow of workers with higher initial wage

to other occupations will tend to decrease wages within the occupation group.

IV) The Effect of Immigration on Labor Market Outcomes

Next, we study the impact of immigration on the wages and employment of natives. We have

provided in the previous section evidence of selective mobility of natives across locations and

occupations in response to immigrant inflows. To assess how this endogenous mobility affects

the estimated wage impact of immigration, we estimate four alternative models which differ

with respect to their inclusion rule in the final period. First, we address concerns about sample

selection in our group of location stayers by carrying an intention-to-treat analysis. In this

specification, we include movers in the sample but assign them the immigrant inflow into

their initial location. Then, we estimate a second model in which individuals who have moved

from the location are excluded. This ensures that our identifying variation in this second

model arises from changes in immigrant inflows for stayers. A comparison between these two

estimates will indicate how much including stayers in the sample affects the results.

A third model is estimated by further restricting the sample to those who remained in

the same occupation group and location during both periods. Finally, we estimate a fourth

model using repeated cross-section of workers in the occupation group across locations.

Workers in the cross-section sample in the initial period are the same as in the balanced

sample, but they differ in the second period given some workers have left and new workers

have entered the occupation group in the location.27

Differences across the estimated models arise through the sorting of individuals across

locations and occupations. A comparison of these estimates will thus indicate how much self-

27 To keep the results comparable, as in the balanced sample, we also focus on the observed change in outcomes for workers aged 25-45 in the initial period and 35-55 in the end period.

26

selection into occupations and location affects the measured impact of immigration on wages.

If changes in occupation and location are not important at the aggregate level, we should find

little differences across models. Instead, if compositional changes are correlated with

immigrant inflows, the estimates should differ.

Effect on the number of days worked

We start by estimating the effect of immigration on the number of days worked. To do so, we

use as a dependent variable the relative change in the number of days worked 1kt kt

kt

D D

D

where

ktD is the average number of days worked by group k in period t.28

The first column in each subpanel of Table 7 presents regression results using a balanced

sample of natives defined by their initial occupation. Overall, we find no evidence of a

correlation between the change in the average number of days worked and an increase in the

share of low-educated migrants. The coefficients of 2SLS models are sometimes negative but

relatively small in most specifications. A notable fact is that we find larger coefficients in the

construction sector: for this group, the estimates suggest that an increase of 10 p.p. of the

immigration rate would decrease by 2% of the average number of days worked. However, the

point estimates are very imprecise and are not statistically different from zero.

A potential risk for the validity of our results is attrition. One drawback of the previous

measure of number of days worked is that individuals supplying zero days of work are

excluded from the DADS sample. Our estimates might thus be affected by a selection bias if

some natives do not work during a year as a result of immigration. To address this limitation,

panel A in Table 8 uses data from a balanced sample of workers in which zero days of work

have been imputed to individuals who are not in the sample in period 1t . These estimates

28 Results are identical if one uses changes in the log of the average number of days worked in the cell.

27

correct for the effect of non-participation in period t+1 under the strong assumption that all

individuals not observed in the sample have been out of the labor force during the year. Once

again, we do not find a negative impact of immigration in these specifications. If anything,

2SLS models report a positive correlation which is inconsistent with the hypothesis that

immigration might decrease the number of days worked of natives.

The previous estimates should be interpreted with caution given there are some doubts

about the quality of the variable on the number of days worked available in the DADS. For

this reason, Panel B in Table 8 reports estimates of the impact of changes in the proportion of

immigrants on the employment to population rate using Census data. As in the previous

section, we estimate separate models for various groups of natives defined by their education

level.29 The general pattern is very similar: changes in immigrant rates do not appear to be

correlated with a decline in employment rate for prime-age male workers from different

education groups. Parameter estimates are always small, and most of the time statistically

insignificant.30

Effect on wages

We next estimate the effect of immigration on wages. We focus on changes in the median

annual log wage for full time workers in a given occupation group. Using the median annual

wage has the advantage of providing estimates relatively insensitive to the presence of outliers

and, in addition, enables a series of robustness checks related to attrition that we present

below. We also explore other options below, however, with specifications that use median

daily wages and average daily wages.

29 Notice that we follow the same sample requirement, and use the change in employment rate of male aged 25-45 in period t and male aged 35-55 in period t+1.

30 We have also estimated similar model by using the initial location of a worker to calculate the employment rates, thus making an “intend to treat” estimate. We also found no effects.

28

The relationship between changes in median annual log wages and immigrant inflows

is examined in Table 9. The OLS coefficients for most groups are rather small, positive and

non- significant in most specifications. In contrast, endogeneity-corrected parameter estimates

significantly differ across occupations. Point estimates are negative, relatively large and

statistically significant. Results with the balanced sample including location-movers indicate

that an increase of 10 p.p. in the share of low-educated immigrants lowers the median log annual

wage by -1.3% for workers initially blue collar in the non-tradable sector. For workers in the

tradable sector, the coefficient is also negative and has the same magnitude but is not

statistically significant. The impact of immigration also appears to be quite heterogeneous

across occupation groups: the results show larger effects for workers in the non-tradable sector

and particularly those in the construction sector. In the construction sector, the estimates

indicate that a 10 p.p. increase predicts a decrease of the median wage by 3.6 p.p. Qualitatively,

these estimates are in line with Bratsberg and Raaum (2012) who report that an increase in 10

p.p. in the immigration ratio is accompanied by a 6 p.p. fall in the wages of construction

workers.

A comparison between Columns 2 and 3 within each panel indicates the extent to

which location and occupation stayers are a selected sample. The estimates present a

contrasting pattern: the estimates in Column 2 where location-movers have been excluded are

quite close for blue collar workers or lower for construction workers but they are slightly

larger for workers in the non-tradable sector. Overall, differences between models including

or excluding location-movers are relatively small for most groups.

In Column 3, where the sample excludes those who are not in the same occupation

group in the second period, the coefficients are unambiguously larger for most groups. The

estimates indicate that an increase of 10 p.p. of the immigration rate predicts a decrease of

29

median wage of 1.3 p.p. for blue collar, 1.7 p.p. for the non-tradable sector and 4.8 p.p. for the

construction sector.

Finally, results in Column 4 are based on the cross-sectional variation of wages within

occupations. The impact of immigration is substantially larger in cross-section estimates

except for the construction sector. With respect to estimates using the balanced sample, the

estimated coefficient is multiplied by two for blue collar workers and non-tradable sector

workers.

There are two main lessons from the previous results. First, the findings described

above point to a strong heterogeneity in the estimated effect of immigration across workers

depending on their initial occupation. Unsurprisingly, the estimates are much larger for

workers initially in the non-tradable sector and in the construction sector.

Second, we find a much larger effect of immigration on those who stay in the skill

group during both periods which implies that occupational mobility mitigates the impact of

immigration, particularly for workers who are initially in the construction sector. This

suggests that immigration affects to a larger extent the average wages in given occupations

than the wages of workers initially in those occupations. Instead, estimates including or

excluding location-movers are basically equivalent, which shows that geographical mobility

does not seem to play an important role in mitigating the impact of immigration.

Robustness

Table 10 examines the sensitivity of the results to the specification of the baseline model. We

examine in Panel 1 the robustness of estimates using the balanced panel while Panel 2 reports

cross-section estimates. We concentrate on the impact of immigration on wages and we focus

on blue collar workers and those in the non-tradable and in the construction sector. We focus

on 2SLS estimates to save space (OLS results are available upon request).

30

We first examine the results for the balanced panel in Panel 1. One issue with our

instruments might be that the lagged distribution of immigrants is correlated with persistent

trends in economic dynamism across locations. As a result, the exclusion restriction of our

estimates might not be perfectly valid. A simple way to test this hypothesis is to estimate

whether the estimates change when we exclude different sets of control variables.31 If the

estimates change significantly, this would indicate that the immigrant inflows predicted by

our instrument are strongly correlated with other factors influencing wages across locations.

Rows 1, 2 and 3 examine the sensitivity of the previous estimates to the inclusion of an

increasingly detailed set of control variables. A comparison between regression results in row

1 which do not includes controls (except for time dummies) and row 2 suggests that results

are barely affected by the inclusion of additional covariates. There is also very little change to

the coefficients from adding regional trends (row 3). Overall, these patterns are not consistent

with the hypothesis that our instrument might be correlated with unobserved determinants of

wage changes across locations.

An additional test for the validity of our instrumental variable strategy is provided by

using more distant lags to compute the instrument. To do so, we construct an alternative

instrument using the distribution of immigrant communities corresponding to two censuses

before.32 By increasing the distance between the initial distribution of immigrants across

locations used to compute the shift share and the change in the immigrant ratio predicted by

the shift share, we are more likely to purge the instrument for potential persistent correlations

with unobserved local trends. Row 4 provides results using this alternative instrument. As

previously, we obtain negative coefficients which are statistically significant. However, these

31 Another good reason to exclude the set of location specific control variables is that these controls might be endogenous. This would be the case for example if variables such as the share of workers in the construction sector or in the manufacturing sector are significantly affected

by immigrant inflows.

32 Attempts to use very distant lags such as predicting changes from 1975 to 2007 by using only the 1968 distribution failed because they are too weakly correlated with changes which occurred during the 1990s and the 2000s. This is due to the fact that more than half low-educated

migrants who arrived after 1980 come from Asia and South-Africa and these groups were quasi-absent from France before 1975.

31

coefficients are also smaller by a third for blue-collar workers and workers in the non-tradable

sector.

The analysis so far has been based on changes in median annual log wages. We now

explore the sensitivity of the results to the definition of the dependent variable. Row 5 uses

the log of the median daily wages in the cell instead as a dependent variable. The estimated

effects are somewhat lower, particularly for workers in the construction sector. Row 6 shows

results using average daily wages. The estimated coefficient is strikingly similar than those

obtained with median daily wages for blue collar workers and is larger for workers in the non-

tradable sector. In contrast, the estimated impact on construction workers diminishes widely

and becomes insignificant. We suspect this last result reflects the fact that a relatively larger

share of construction workers has several employers in a given year. This implies that there

might be much more measurement errors in the number of days worked for this group when

daily wages are used.

In Rows 7 and 8, we investigate the extent to which the results might be driven by

large locations such as Paris, Marseilles or Lyons, which attract a disproportionately large

share of immigrants. Row 7 presents estimates where the 30 largest commuting zones have

been excluded from the sample while row 8 reports unweighted regressions. Results are

broadly similar in these two models.

The final specification check in Row 9 is oriented towards addressing a number of

concerns related to attrition. Using median wages in the cell enables us to investigate the

sensitivity of the results to missing wage information by using simple imputation techniques

as in Neal and Johnson (1996) or Olivetti and Petrongolo (2008). Specifically, we investigate

how our results depend on including individuals with a missing wage observation in period

1t in the sample under the assumption that all missing individuals are out of the labor force

and are thus earning a log wage of zero. Results of this exercise in Row 9 indicate our

32

estimates are reasonably robust. We also obtain a similar ranking across occupations, point

estimates being slightly larger for workers from non-tradable industries and lower for the blue

collar group.33

Panel B reports the same robustness tests performed using the cross-section sample.

We also find the baseline results to be reasonably robust across most specifications but the

precision of the estimates diminishes in some specifications. Importantly, the estimates are

not statistically significant when the alternative instrument using lagged settlement patterns of

immigrants in used. Another noteworthy pattern is that unweighted specifications provide

much larger parameter estimates.

V) Discussion

In this paper, we have revisited the impact of immigration on the labor market outcomes of

natives. Unlike most of the previous literature, our rich dataset has provided us with a unique

opportunity to investigate heterogeneity in the impact of immigration while controlling for

composition effects. Specifically, we have tested whether the impact of low-educated

immigration differed across natives using homogenous groups defined by their initial

occupation. We have also controlled for changes in the composition of the labor force at the

local level by focusing on estimates using variations over time from a balanced sample.

First, our findings show that immigrant inflows are moderately correlated with both

native outflows and inflows, and with a reallocation of natives to occupations with less routine

tasks. Our results also point to the evidence that the correlation between immigrant inflows

mobility across locations and occupations varies strongly depending on the industry of origin.

Moreover, we find that location and occupation-movers are a selected subgroup of the sending

33 We have also evaluated the risk of attrition from a selective shift of some workers to a sector uncovered by the DADS panel such as the public sector which is only partially covered by the DADS before 1990. We found basically no correlations between changes in the share of

government employees and a change in the share of migrants. These results are available upon request.

33

population. While location-movers tend to be negatively selected from the sending population,

those moving to occupations with less routine tasks tend to be positively selected. This implies

that the selection patterns related to changes in location and occupation strongly differ.

Empirically, we do not find strong evidence that the selection of natives across locations

affects importantly the estimates of the impact of immigration. On the other hand, the

endogenous selection of natives towards occupations requiring different skills decreases

significantly the impact of immigration on wages. We obtain a much larger impact of immigrant

inflows on wages on the selected subgroups of natives who do not change location and stay in

the same occupation group in both periods.

Importantly, our results point to a strong heterogeneity across occupation groups in the

impact of immigration. We find that the wages of blue collar workers initially in the tradable

sector are the less affected by immigrant inflows while the effect of immigration on changes in

occupations are found stronger for this group. In contrast, for the group of construction workers,

we find a much larger impact of immigration on wages but no correlation between immigrant

inflows and change in occupations for this group. This last result suggests that part of the

difference in the impact of immigration on wages across occupation groups might reflect the

fact that workers in some occupation groups are more able than others to protect themselves

from immigrant inflows by shifting occupation.

Our results have important implications for the analysis of the impact of immigration.

First, the results suggest that the wage impact of immigration varies across occupation groups

and individuals both because labor markets are segmented and also because natives between

and within groups react differently to immigrant inflows. Second, the fact that immigrant

inflows are correlated with native mobility across location and occupation groups complicates

the estimates of the impact of immigration. Native mobility responses imply that immigration

affects indirectly other locations and occupation groups.

34

There are however several limitations to the previous results. First, because we wanted

to minimize the risk that our results might be biased by non-participation which is not well

captured in our data, we have focused on prime aged male workers. According to recent work

from Smith (2012), immigrants might also be more in competition with young workers less

than 25 that were not included in our analysis. Another limitation is that we did not include

women in our analysis given that the treatment of labor market participation creates an

additional complexity for this group. Finally, the share of immigrants and natives in the service

sector is also rapidly growing while the share of blue collar workers is in constant decline. An

evaluation of the impact of immigration on this increasingly important segment of the labor

market would be of substantial interest for future work.

Appendix

Data appendix

Occupations: DADS data contains information on occupation for CSP with 27 categories

before 1983 and 36 categories afterward. The category “Blue collar workers” aggregates 6

distinct sub-occupations over the period. We merge these occupations with tasks intensity

indexes by Goos et al. (2010, Table 4 p.49) based on the Occupational Information Network

(ONET) database.

Crosswalk tables for industry classifications: We use the industry classification which

remained unchanged for the longest period of time in the data. The NAP (Nomenclatures

d'Activités et de Produits 1973) is used in the 1975, 1982 and 1990 censuses and in the DADS

until 1993. We have created crosswalk tables with other industry classifications to match

them with the NAP at the four digit level. The NAF (Nomenclature d'Activité Française) is

used in the 1999 Census and in the DADS from 1993 to 2002. For the match between NAP

and NAF, we have used the 1994 LFS (Enquête emploi) in which both codes are also given to

establish a match at the four digit levels. Similarly, when several possibilities existed, we have

35

kept the most frequent correspondence. In both cases, the match has been completed manually

to include exhaustively all codes in the correspondence table at the four digit level.

Education The education variable reported in the Census indicates the diploma received by

the individual. We use the variable DIP in the 1968, 1975 and 1982 censuses, DIPL1 in the

1990 Census and DIPL in the 1999 Census. We classify individuals in four groups: Primary

education, Secondary education, High School and College. Primary education level includes

individuals which declare to have no diploma and people having the primary school

certificates. Secondary education level includes individuals which report to have a diploma of

a level equivalent to the Diplôme National du Brevet (BEPC) and includes individuals holding

a CAP or a BEP. High school education includes individuals who have a diploma equivalent

to the Baccalaureate. This group also includes general, professional or technical Baccalaureate

graduates. College level includes all individuals with a diploma of a level superior to the

Baccalaureate.

Theoretical Appendix

The model is a straightforward adaptation of Combes et al. (2008) and Combes et al. (2012)

which was initially used to investigate the sorting of workers with different skill levels across

local labor markets. We also follow Card (2001) and make the assumption that local labor

markets are stratified along “skill” group lines. For the moment, we abstract from labor

supply decisions and assume that each worker provides one unit of labor. This implies that

local labor supply is only determined by workers’ location decisions. The profit of the

representative firm in location l , and year t is given by:

lt lt lt it it lt lt

i lt

wp y l r z

36

In this expression, ltp is the local price of output lty . For any worker i , itw is the wage rate

and itl is the total labor supply from workers of this type. Other factors of production are

represented by ltz and ltr is their price. The production function of the firm is:

1b b

lt lt lt lty A L z

where 0 1b , ltA is the total factor productivity in the area. We assume ltL is a CES

aggregate of the labor inputs of workers from different occupation groups. As in Combes

et al. (2008), to introduce heterogeneity within occupation groups, we assume individuals in a

given occupation group are perfect substitute but supply different efficiency units of labor.

This implies that:

/ 1/ 1 ( 1)/

( 1)/

lt klt it it

i kk k lt

L L s l

where determines the elasticity of substitution between labor types. The level of skills its

augments the effective units of labor supplied by a given individual, thereby making workers

more productive. At competitive equilibrium, profit maximization implies:

1/( , )it lt it kltw k l B s L

where 1 1

(1 ) 1/( 1)(1 )b

bb blt lt lt lt ltB b b p A r L

. To take the model to the data, we assume that for

workers of type i : log it it is X , where itX is a vector of time-varying observed

characteristics of the worker and i is a worker fixed effect unobserved productivity.

Firms can employ native labor kltN or immigrant labor

ltI , we assume that immigrant

and native labor of the same type are perfect substitute. Labor supply in an occupation group

can thus be decomposed by the efficiency units supplied by natives and low-educated

37

immigrants. Finally, using the following approximation, log log ltklt klt k

klt

IL N

N , it is

straightforward to obtain Eq. (1).

References

Abowd, John M., Francis Kramarz and David N. Margolis (1999), ‘High wage workers and

high wage firms’, Econometrica 67(2), 251–334.

Acemoglu, Daron and David. Autor (2011), Skills, tasks and technologies: Implications for

employment and earnings, in ‘Handbook of Labor Economics’, Vol. 4.

Aeberhardt, Romain, Pauline Givord and Claire Marbot (2011), ‘Minimum wage and wage

inequality in France: An unconditional quantile regression approach’.

Amuedo-Dorantes, Catalina and Sara de la Rica (2011), ‘Complements or substitutes? task

specialization by gender and nativity in Spain’, Labour Economics 18(5), 697 – 707.

Autor, David H., Frank Levy and Richard J. Murnane (2003), ‘The skill content of recent

technological change: An empirical exploration’, The Quarterly Journal of Economics

118(4), 1279–1333.

Aydemir, Abdurrahman and George J. Borjas (2007), ‘Cross-Country Variation in the Impact

of International Migration: Canada, Mexico, and the United States’, Journal of the

European Economic Association 5(4), 663–708.

Aydemir, Abdurrahman and George J. Borjas (2011), ‘Attenuation bias in measuring the

wage impact of immigration’, Journal of Labor Economics 29(1), 69–112.

Beaudry, Paul, Mark Doms and Ethan Lewis (2010), ‘Should the personal computer be

considered a technological revolution? evidence from u.s. metropolitan areas’, Journal

of Political Economy 118(5), 988 – 1036.

Borjas, George J (1999), ‘The economic analysis of immigration’, Handbook of labor

economics 3, 1697–1760.

Borjas, George J. (2003), ‘The Labor Demand Curve is Downward Sloping: Reexamining the

Impact of Immigration on the Labor Market’, Quarterly Journal of Economics

118(4), 1335–1374.

Borjas, George J. (2006), ‘Native Internal Migration and the Labor Market Impact of

Immigration’, Journal of Human Resources 41(2), 221.

38

Borjas, George J., Richard B. Freeman and Lawrence F. Katz (1997), ‘How much do

immigration and trade affect labor market outcomes?’, Brookings Papers on Economic

Activity (1), 1–90.

Bratsberg, Bernt and Oddbjørn Raaum (2012), ‘Immigration and wages: Evidence from

construction’, The Economic Journal, 122(1), 1177–1205.

Bratsberg, Bernt, Oddbjørn Raaum, Marianne Røed and Pål Schøne (2014), ‘Immigration

Wage Effects by Origin’, Scandinavian Journal of Economics . 116(2), 356-393.

Card, David (2001), ‘Immigrant inflows, native outflows, and the local market impacts of

higher immigration’, Journal of Labor Economics 19(1), 22–64.

Card, David (2009), ‘Immigration and inequality’, American Economic Review (Papers and

Proceedings) 99(2), 1–21.

Card, David (2012), ‘Comment: The elusive search for negative wage impacts of

immigration’, Journal of the European Economic Association 10(1), 211–215.

Card, David and John DiNardo (2000), ‘Do immigrant inflows lead to native outflows?’, The

American Economic Review 90(2), 360–367.

Combes, Pierre-Philippe, Gilles Duranton and Laurent Gobillon (2008), ‘Spatial wage

disparities: Sorting matters!’, Journal of Urban Economics 63(2), 723–742.

Combes, Pierre-Philippe, Gilles Duranton, Laurent Gobillon and Sébastien Roux (2012),

‘Sorting and local wage and skill distributions in France’, Regional Science and Urban

Economics . 42(6), 913-930.

Cortes, Patricia (2008), ‘The effect of low-skilled immigration on u.s. prices: Evidence from

CPI data’, Journal of Political Economy 116(3), 381–422

De New, John P. and Klaus F. Zimmermann (1994), ‘Native wage impacts of foreign labor: a

random effects panel analysis’, Journal of Population Economics 7, 177–192.

Dustmann, Christian and Albrecht Glitz (2015), ‘How do industries and firms respond to

changes in local labor supply’, Journal of Labor Economics 33(3).

Dustmann, Christian, Tommaso Frattini and Ian Preston (2013), ‘The effect of immigration

along the distribution of wages’, The Review of Economic Studies 80(1), 145 – 173.

Foged, Mette and Giovanni Peri (2014), Immigrants and Native Workers: New Analysis

Using Longitudinal Employer-Employee Data, Mimeo.

Goos, Maarten and Alan Manning (2007), ‘Lousy and lovely jobs: The rising polarization of

work in Britain’, The Review of Economics and Statistics 89(1), 118–133.

Hanson, Gordon and Matthew J. Slaughter (2002), ‘Labor-market adjustment in open

economies: Evidence from us states’, Journal of International Economics 57(1), 3–29.

39

Hatton, Timothy J. and Massimiliano Tani (2005), ‘Immigration and inter-regional mobility

in the uk, 1982-2000’, Economic Journal 115(507), F342–F358.

Hunt, Jennifer (2012), The impact of immigration on the educational attainment of natives,

Working Paper 18047, National Bureau of Economic Research.

Hunt, Jennifer and Marjolaine Gauthier-Loiselle (2010), ‘How much does immigration boost

innovation?’, American Economic Journal: Macroeconomics 2(2), 31–56.

Koubi, Malik and Sébastien Roux (2004), Refonte du panel DADS : principes et premières

estimations d’emploi et de salaire. Version provisoire, note interne 204/F240, Insee.

Llull, Joan (2014), Immigration, Wages, and Education: A Labor Market Equilibrium

Structural Model, Mimeo.

Manacorda, Marco, Alan Manning and Jonathan Wadsworth (2012), ‘The Impact of

Immigration on the Structure of Wages: Theory and Evidence from Britain’, Journal of

the European Economic Association 10(1), 120–151.

Mazzolari, F. and D. Neumark (2012), ‘Immigration and product diversity’, Journal of

Population Economics, 25(3), 1107-1137.

Mocetti, Sauro and Carmine Porello (2010), ‘How does immigration affect native internal

mobility? new evidence from Italy’, Regional Science and Urban Economics

40(6), 427–439.

Moretti, Enrico (2011), Local Labor Markets, Vol. 4 of Handbook of Labor Economics,

Elsevier, chapter 14, pp. 1237–1313.

Neal, Derek A. and William R. Johnson (1996), ‘The role of premarket factors in black-white

wage differences’, Journal of Political Economy 104(5), 869–95.

Olivetti, Claudia and Barbara Petrongolo (2008), ‘Unequal pay or unequal employment? a

cross-country analysis of gender gaps.’, Journal of Labor Economics 26(4), 621 – 654.

Ortega, Javier and Gregory Verdugo (2014), ‘The impact of immigration on the French labor

market: why so different?’, Labour Economics 29, 14-27.

Ottaviano, Gianmarco I.P. and Giovanni Peri (2012), ‘Rethinking the effect of immigration on

wages’, Journal of the European Economic Association 10(1), 152–198.

Peri, Giovanni and Chad Sparber (2009), ‘Task specialization, comparative advantages, and

the effects of immigration on wages’, American Economic Journal: Applied Economics

1(3), 135–169.

Royer, Jean-François (2007), ‘Quatre observations sur la mobilité résidentielle en France

métropolitaine’, Série des documents de travail du CREST (10).

40

Smith, Christopher L. (2012), ‘The impact of low-skilled immigration on the youth labor

market’, Journal of Labor Economics 30(1), 55 – 89.

Wozniak, Abigail and Thomas J. Murray (2012), ‘Timing is everything: Short-run population

impacts of immigration in us cities’, Journal of Urban Economics 72(1), 60 – 78.

41

Table 1 : Share of Foreign born Workers among blue collar workers

across Selected Industries and Regions in France in 1999

Industry Share of foreign born workers Share total

Employment

France Paris Lyons Brittany

Non-Tradable 14.5 62.0 16.2 3.4 62.0

Tradable 10.6 31.1 15.7 1.9 31.1

Construction 20.2 16.4 23.9 5.0 16.4

Source: Panel DADS. All figures refer to blue collar workers only. Paris and Lyons regions

refer respectively to the region “Ile de France” and “Rhone-Alpes”.

Table 2: Share of Foreign Born among Construction Workers

in the Paris and Brittany regions, 1976-2007

1976 1982 1990 1999 2007

Paris 37.2 37.1 35.6 45.7 41.7

Brittany 3.8 4.2 4 5.0 6.9

Source: Panel DADS. All figures refer to blue collar workers.

Table 3: First Stage Results

Dependent variable : Change in Low-Educated

Immigrant Ratio ltp

(1) (2) (3)

Predicted change 0.166*** 0.121*** 0.091***

(0.051) (0.028) (0.027)

F-stat 10.54 17.74 10.97

R-squared 0.13 0.33 0.29

Additional Controls No Yes Yes

Weight Yes Yes No

Note: All regressions use 1188 observations and include a full set of regions and time fixed

effects. Additional controls included in the regressions when indicated. Standard errors are

clustered at the commuting zone level. Regressions are weighted by 1/2

1(1/ 1/ )klt kltN N

except when indicated otherwise. A (*) denotes statistical significance at the 10% level, a (**)

denotes at the 5% level, a (***) at the 1% level.

42

Table 4: Impact of Low-Educated Immigration on

Native inflows and outflows at the Commuting Zone Level

1. DADS Data

Sample: Male workers 25-45 in t, 35-55 in 1t ,

initially working in industry.

All

Blue Collar

Tradable Non-

tradable

Construction

A. Dependent variable: Outflows between

t/t+1

OLS

ltp 0.070*** 0.046 0.089*** 0.090***

(0.024) (0.028) (0.030) (0.030)

2SLS

ltp 0.160* 0.005 0.248** 0.358**

(0.091) (0.118) (0.122) (0.154)

B. Dependent variable: Inflows between t/t+1

OLS

ltp 0.160*** 0.156*** 0.164*** 0.173***

(0.031) (0.039) (0.036) (0.050)

2SLS

ltp 0.147 0.165 0.144 0.158

(0.104) (0.114) (0.138) (0.169)

2. Census Data

Primary

Education

Secondary

Education

High-

School University

A. Dependent variable: Outflows between t/t+1

OLS

ltp 0.044*** 0.043** -0.029 -0.027

(0.014) (0.019) (0.021) (0.029)

2SLS

ltp 0.238** 0.274** 0.238 0.125

(0.096) (0.131) (0.178) (0.202)

B. Dependent variable: Inflows between t/t+1

OLS

ltp 0.110*** 0.137** 0.104 0.104

(0.041) (0.057) (0.074) (0.109)

2SLS

ltp 0.222 0.258 0.456 0.560

(0.176) (0.230) (0.357) (0.514)

Note: All regressions use 1188 observations and include a full set of regions and time fixed

effects. Additional controls included in the regressions. Standard errors are clustered at the

43

commuting zone level. Regressions are weighted by 1/2

1(1/ 1/ )klt kltN N

. A (*) denotes

statistical significance at the 10% level, a (**) denotes at the 5% level, a (***) at the 1% level.

44

Table 5: Impact of Immigration on Occupation Characteristics A. Dependent variable : change in average routine task t/t+1

Sample: Male 25-45 in t+1, 35-55 in t

Blue Collar Tradable

Balanced

sample

Location

Stayer

Location &

Occupation

Stayer

Cross-

Section

Balanced

sample

Location

Stayer

Location &

Occupation

Stayer

Cross-

Section

OLS

ltp 0.03 0.023 0.017 0.010 0.003 0.000 0.028** 0.025**

(0.039 (0.050) (0.012) (0.013) (0.054) (0.069) (0.014) (0.011)

2SLS

ltp -0.623*** -0.653*** -0.147** -0.049 -0.791*** -0.703** -0.137* -0.199**

(0.184) (0.199) (0.072) (0.079) (0.270) (0.245) (0.075) (0.090)

Non-tradable Construction

OLS

ltp 0.052 0.053 -0.001 0.005 0.028 0.008 -0.020* -0.016

(0.040) (0.054) (0.014) (0.013) (0.039) (0.053) (0.012) (0.012)

2SLS

ltp -0.388** -0.434** -0.041 0.051 -0.181 -0.370 -0.02 -0.011

(0.168) (0.214) (0.080) (0.092) (0.197) (0.293) (0.082) (0.063)

Note: All regressions use 1188 observations and include a full set of regions and time fixed

effects. Additional controls are included in the regressions. Standard errors are clustered at the

commuting zone level. Regressions are weighted by 1/2

1(1/ 1/ )klt kltN N

. A (*) denotes

statistical significance at the 10% level, a (**) denotes at the 5% level, a (***) at the 1% level.

45

Table 6: Selection Patterns of Location and Occupation-Movers:

Individual Level Evidence

A. Dependent Variable: Moved to another Location

Blue Collar Tradable Non-Tradable Construction

Residual Wage -0.117 -0.119 -0.089 -0.081 -0.098 -0.105 -0.094 -0.116

(0.003) (0.006) (0.007) (0.017) (0.003) (0.007) (0.007) (0.018)

RW x ltp 0.018 -0.114 0.101 0.287

(0.085) (0.224) (0.116) (0.297)

ltp 0.258 0.257 0.263 0.264 0.249 0.245 0.545 0.518

(0.091) (0.094) (0.115) (0.118) (0.095) (0.098) (0.116) (0.121)

N 345 414 345 414 157 569 157 569 187 841 187 841 60 609 60 609

B. Dependent variable: Change in routine to abstract task intensity,

excludes location-movers

Residual Wage -0.355 -0.361 -0.578 -0.585 -0.232 -0.227 -0.315 -0.345

(0.010) (0.017) (0.019) (0.039) (0.011) (0.014) (0.018) (0.055)

RW x ltp 0.090 0.118 -0.077 0.436

(0.195) (0.477) (0.240) (0.821)

ltp -0.502 -0.507 -0.369 -0.372 -0.377 -0.373 -0.455 -0.488

(0.182) (0.187) (0.153) (0.154) (0.173) (0.180) (0.367) (0.351)

N 275 854 275 854 134 977 134 977 140 705 140 705 46 945 46 945

Note: All regressions include a full set of regions and time fixed effects. Additional controls

are included in the regressions. Standard errors are clustered at the regional level. All models

are estimated with 2SLS.

46

Table 7: Impact of immigration on Number of Days Worked Dependent variable : change in average number of days worked t/t+1

Sample: Male 25-45 in t, 35-55 in t+1

Blue Collar Tradable

Balanced

sample

Location

Stayer

Location &

Occupation

Stayer

Cross-

Section

Balanced

sample

Location

Stayer

Location &

Occupation

Stayer

Cross-

Section

OLS

ltp

0.007 -0.003 -0.002 0.009 0.001 0.009 -0.000 0.019

(0.007) (0.006) (0.006) (0.010) (0.008) (0.012) (0.008) (0.012)

2SLS

ltp

0.013 -0.003 -0.015 -0.079 0.052 0.061 0.041 -0.047

(0.034) (0.027) (0.035) (0.068) (0.042) (0.062) (0.051) (0.073)

Non-tradable Construction

OLS

ltp

0.011 -0.009 -0.010 -0.003 0.003 0.031 -0.013 0.000

(0.010) (0.009) (0.009) (0.012) (0.013) (0.027) (0.011) (0.016)

2SLS

ltp

0.011 -0.042 -0.024 -0.083 -0.148 -0.267 -0.232 -0.214

(0.010) (0.042) (0.055) (0.085) (0.105) (0.178) (0.211) (0.151)

Note: All regressions use 1188 observations and include a full set of regions and time fixed

effects. Additional controls are also included in the regressions. Standard errors are clustered

at the commuting zone level. Regressions are weighted by 1/2

1(1/ 1/ )klt kltN N

. A (*)

denotes statistical significance at the 10% level, a (**) denotes at the 5% level, a (***) at the

1% level.

47

Table 8: Additional Evidence on the Impact of Immigration on Number of Days Worked

Dependent variable: change in average number of days worked t/t+1

Sample definition: Male 25-45 in t, 35-55 in t+1

A. Balanced Sample : Zero imputed if missing in t+1

Blue Collar Tradable Non-Tradable Construction

OLS

ltp -0.037 -0.035 -0.044 -0.027

(0.023) (0.021) (0.027) (0.028)

2SLS

ltp 0.328** 0.329 0.219* 0.280*

(0.148) (0.196) (0.124) (0.165)

B. Census Data Evidence:

Dependent Variable: Change in Employment/Population Rate t/t+1

Primary

Education

Secondary

Education High-School University Graduates

OLS

ltp 0.026** 0.009* 0.003 -0.001

(0.011) (0.005) (0.008) (0.046)

2SLS

ltp -0.074 0.004 -0.031 -0.030

(0.095) (0.047) (0.046) (0.046)

Note: All regressions use 1188 observations and include a full set of regions and time fixed

effects. Additional controls are also included in the regressions. Standard errors are clustered

at the commuting zone level. Regressions are weighted by 1/2

1(1/ 1/ )klt kltN N

. A (*)

denotes statistical significance at the 10% level, a (**) denotes at the 5% level, a (***) at the

1% level.

48

Table 9: Impact of Immigration on Median Annual Wages Dependent variable : change in median annual wage t/t+1

Sample: Male 25-45 in t+1, 35-55 in t

Blue Collar Tradable

Balanced

sample

Location

Stayer

Location &

Occupation

Stayer

Cross-

Section

Balanced

sample

Location

Stayer

Location &

Occupation

Stayer

Cross-

Section

OLS

ltp

0.015 0.016 0.030 0.042** 0.012 0.035 0.012 0.059**

(0.013) (0.013) (0.012) (0.019) (0.017) (0.022) (0.016) (0.023)

2SLS

ltp

-0.131** -0.107* -0.130* -0.285* -0.126 0.039 -0.185** -0.235

(0.057) (0.060) (0.073) (0.163) (0.086) (0.100) (0.087) (0.166)

Non-tradable Construction

OLS

ltp

0.008 0.016 0.019 0.007 0.022 0.031 0.015 0.027

(0.019) (0.016) (0.016) (0.020) (0.024) (0.026) (0.029) (0.025)

2SLS

ltp

-0.133** -0.162* -0.171* -0.361** -0.367** -0.267* -0.479** -0.467**

(0.065) (0.087) (0.095 (0.166 (0.17 (0.154) (0.240) (0.199)

Note: All regressions use 1188 observations and include a full set of regions and time fixed

effects. Additional controls are included in the regressions. Standard errors are clustered at the

commuting zone level. Regressions are weighted by 1/2

1(1/ 1/ )klt kltN N

where kltN

represents the size of the occupation group k in location l and year t . A (*) denotes

statistical significance at the 10% level, a (**) denotes at the 5% level, a (***) at the 1% level.

49

Table 10: Sensitivity of the Effects of Immigration

on Median wages to alternative specifications

A. Balanced Panel

Blue Collar

Non-tradable Construction N

1. No covariates -0.178*** -0.177** -0.354** 1188

(0.069) (0.081) (0.153)

2. Covariates -0.177** -0.174** -0.331** 1188

(0.072) (0.081) (0.153)

3. Only Region FE -0.139** -0.154* -0.416** 1188

(0.058) (0.087) (0.152)

4. Instrument lagged -0.094** -0.091* -0.366** 1188

(0.042) (0.054) (0.106)

5. Dependent variable: -0.170*** -0.112* -0.206** 1188

Median daily wage (0.063) (0.060) (0.130)

6. Dependent variable: -0.170** -0.223** -0.077 1188

Average daily wage (0.080) (0.103) (0.189)

7. Exclude largest cities -0.178** -0.141 -0.419* 1064

(0.090) (0.127) (0.255)

8. Without weights -0.212** -0.155 -0.399* 1188

(0.109) (0.106) (0.222)

9. Log wage of -0.095* -0.195** -0.351** 1188

zero imputed if missing in t+1 (0.055) (0.091) (0.179)

B. Repeated Cross-section

1. No covariates -0.221 -0.243* -0.378* 1188

(0.133) (0.146) (0.201)

2. Covariates -0.212 -0.246* -0.388* 1188

(0.135) (0.149) (0.210)

3. Only Region FE -0.283* -0.374** -0.467** 1188

(0.161) (0.166) (0.196)

4. Instrument lagged -0.092 -0.108 -0.116 1188

(0.080) (0.093) (0.103)

5. Dependent variable: -0.135* -0.154* -0.178 1188

Median daily wage (0.079) (0.080) (0.144)

6. Dependent variable: -0.121 -0.195 -0.295 1188

Average daily wage (0.108) (0.132) (0.196)

7. Exclude largest cities -0.426 -0.459* -0.519* 1067

(0.276) (0.264) (0.307)

8. Without weights -0.670* -0.579** -0.483** 1188

(0.397) (0.282) (0.221)

Note: See Table 9.